Does your CHECKDB hurt, Argenis?

There is a classic blog series by Paul Randal [blog|twitter] called “CHECKDB From Every Angle” which is pretty much mandatory reading for anybody who’s even remotely considering going for the MCM certification, or its replacement (the Microsoft Certified Solutions Master: Data Platform – makes my fingers hurt just from typing it). Of particular interest is the post “Consistency Options for a VLDB” – on it, Paul provides solid, timeless advice (I use the word “timeless” because it was written in 2007, and it all applies today!) on how to perform checks on very large databases.

Well, here I was trying to figure out how to make CHECKDB run faster on a restored copy of one of our databases, which happens to exceed 7TB in size. The whole thing was taking several days on multiple systems, regardless of the storage used – SAS, SATA or even SSD…and I actually didn’t pay much attention to how long it was taking, or even bothered to look at the reasons why - as long as it was finishing okay and found no consistency errors.

Yes – I know. That was a huge mistake, as corruption found in a database several days after taking place could only allow for further spread of the corruption – and potentially large data loss.

In the last two weeks I increased my attention towards this problem, as we noticed that CHECKDB was taking EVEN LONGER on brand new all-flash storage in the SAN! I couldn’t really explain it, and were almost ready to blame the storage vendor. The vendor told us that they could initially see the server driving decent I/O – around 450Mb/sec, and then it would settle at a very slow rate of 10Mb/sec or so. “Hum”, I thought – “CHECKDB is just not pushing the I/O subsystem hard enough”. Perfmon confirmed the vendor’s observations.

Dreaded @BlobEater

What was CHECKDB doing all the time while doing so little I/O? Eating Blobs.

It turns out that CHECKDB was taking an extremely long time on one of our frankentables, which happens to be have 35 billion rows (yup, with a b) and sucks up several terabytes of space in the database. We do have a project ongoing to purge/split/partition this table, so it’s just a matter of time before we deal with it.

But the reality today is that CHECKDB is coming to a screeching halt in performance when dealing with this particular table.

Checking sys.dm_os_waiting_tasks and sys.dm_os_latch_stats showed that LATCH_EX (DBCC_OBJECT_METADATA) was by far the top wait type. I remembered hearing recently about that wait from another post that Paul Randal made, but that was related to computed-column indexes, and in fact, Paul himself reminded me of his article via twitter. But alas, our pathologic table had no non-clustered indexes on computed columns.

I knew that latches are used by the database engine to do internal synchronization – but how could I help speed this up? After all, this is stuff that doesn’t have a lot of knobs to tweak.

(There’s a fantastic level 500 talk by Bob Ward from Microsoft CSS [blog|twitter] called “Inside SQL Server Latches” given at PASS 2010 – and you can check it out here. DISCLAIMER: I assume no responsibility for any brain melting that might ensue from watching Bob’s talk!)

Failed Hypotheses

Earlier on this week I flew down to Palo Alto, CA, to visit our Headquarters – and after having a great time with my Monkey peers, I was relaxing on the plane back to Seattle watching a great talk by SQL Server MVP and fellow MCM Maciej Pilecki [twitter] called “Masterclass: A Day in the Life of a Database Transaction” where he discusses many different topics related to transaction management inside SQL Server. Very good stuff, and when I got home it was a little late – that slow DBCC CHECKDB that I had been dealing with was way in the back of my head.

As I was looking at the problem at hand earlier on this week, I thought “How about I set the database to read-only?” I remembered one of the things Maciej had (jokingly) said in his talk: “if you don’t want locking and blocking, set the database to read-only” (or something to that effect, pardon my loose memory). I immediately killed the CHECKDB which had been running painfully for days, and set the database to read-only mode. Then I ran DBCC CHECKDB against it. It started going really fast (even a bit faster than before), and then throttled down again to around 10Mb/sec. All sorts of expletives went through my head at the time. Sure enough, the same latching scenario was present. Oh well.

I even spent some time trying to figure out if NUMA was hurting performance. Folks on Twitter made suggestions in this regard (thanks, Lonny! [twitter])

…Eureka?

This past Friday I was still scratching my head about the whole thing; I was ready to start profiling with XPERF to see if I could figure out which part of the engine was to blame and then get Microsoft to look at the evidence.

After getting a bunch of good news I’ll blog about separately, I sat down for a figurative smack down with CHECKDB before the weekend. And then the light bulb went on.

A sparse column. I thought that I couldn’t possibly be experiencing the same scenario that Paul blogged about back in March showing extreme latching with non-clustered indexes on computed columns. Did I even have a non-clustered index on my sparse column?

As it turns out, I did. I had one filtered non-clustered index – with the sparse column as the index key (and only column).

To prove that this was the problem, I went and setup a test.

Yup, that'll do it

The repro is very simple for this issue: I tested it on the latest public builds of SQL Server 2008 R2 SP2 (CU6) and SQL Server 2012 SP1 (CU4).

First, create a test database and a test table, which only needs to contain a sparse column:

CREATE DATABASE SparseColTest;
GO

USE SparseColTest;
GO

CREATE TABLE testTable (testCol smalldatetime SPARSE NULL);
GO

INSERT INTO testTable (testCol) 
    VALUES (NULL);
GO 1000000

That’s 1 million rows, and even though you’re inserting NULLs, that’s going to take a while. In my laptop, it took 3 minutes and 31 seconds.

Next, we run DBCC CHECKDB against the database:

DBCC CHECKDB('SparseColTest') WITH NO_INFOMSGS, ALL_ERRORMSGS;

This runs extremely fast, as least on my test rig – 198 milliseconds.

Now let’s create a filtered non-clustered index on the sparse column:

CREATE NONCLUSTERED INDEX [badBadIndex]
    ON testTable (testCol)
    WHERE testCol IS NOT NULL;

With the index in place now, let’s run DBCC CHECKDB one more time:

DBCC CHECKDB('SparseColTest') WITH NO_INFOMSGS, ALL_ERRORMSGS;

In my test system this statement completed in 11433 milliseconds. 11.43 full seconds. Quite the jump from 198 milliseconds.

I went ahead and dropped the filtered non-clustered indexes on the restored copy of our production database, and ran CHECKDB against that. We went down from 7+ days to 19 hours and 20 minutes.

Cue the “Argenis is not impressed” meme, please, Mr. LaRock.

My pain is your gain, folks. Go check to see if you have any of such indexes – they’re likely causing your consistency checks to run very, very slow.

Happy CHECKDBing,

-Argenis

ps: I plan to file a Connect item for this issue – I consider it a pretty serious bug in the engine. After all, filtered indexes were invented BECAUSE of the sparse column feature – and it makes a lot of sense to use them together. Watch this space and my twitter timeline for a link.

Retaking on this blog series after an extended hiatus. I hope you will enjoy today’s topic on documentation.

Boring Work? Nope – An Opportunity

One of the often neglected and yet extremely important tasks of any DBA is to document his/her environment. I can certainly relate to those of you who haven’t written a single page of documentation for your environment – I was there before. And I felt bad.

The good news is that documentation doesn’t have to be painful. My friends at Brent Ozar Unlimited recently posted a video on documentation – you can find it in this link – and they touch on that very same subject: There ARE ways to document your environment that will leave you greatly satisfied with the results, and some of you might even enjoy the process. Your manager will love you for it – and you won’t feel bad for the new guy who replaces you when it’s time to move on.

How About Word Documents and Excel Sheets in SharePoint?

Not perfect. Let’s face it. Writing pages and pages of documentation for your database environment (or ANY IT environment) in Word is tiresome. Excel sheets can have tons of information and might be easy to peruse. But no matter how good your formatting is, how many pictures and Visio diagrams you throw at it, you’ll always get the sense that the document you spent days working on will just end up in the SharePoint site and collect dust there. SharePoint can handle changes and versioning for you, if you use it correctly – so that’s a plus. But think about ease of access to the documents – ease of editing. Discussions around it. Even automated updates! Is a SharePoint folder the way to go for that? (Or a UNC path, for those of you without SharePoint?)

An Old Friend Can Work Marvels

You’ve been to WikiPedia. More than once. You know how useful it is, and how, thanks for the built-in search function and Search Engines, easy it is to find information on it.

How about using a wiki engine for your documentation? Can’t picture that in your head? Hear me out: wikis are VERY easy to create, edit and collaborate on – you just click on Edit and away you go. They have built-in version control, discussion pages, categories, lists, and pretty much everything you need to have a versatile documentation site.

Picture this: a wiki page for every environment, listing all the servers that belong to it – and linking to individual wiki pages with tons of information about those servers, including former incident highlights, detailed information about the server (CPUs, RAM, OS Version, SQL Instances, patches, team(s) who own the server, escalation paths for incidents, and much more.

The best part: these pages that have lots of information can be updated via an automated process. WHAT? AUTOMATED DOCUMENTATION YOU SAY? Yes. Think scripts. Scripts that will go over your environments on a regular basis and update information on them in the wiki. Does that sound cool or what? All you need is to understand the capabilities of your wiki engine (think API), and pair it up with your favorite scripting language. For me, it’s a slam-dunk: I’ll use PowerShell to automate the information gathering process, and a wiki engine that runs on SQL Server – I’ll show you a few options in a bit.

Now imagine that you give access to your NOC/Helpdesk to this wiki site: you don’t want them to change anything on it, so you set permissions for read-only access to the site. The benefits would be great: now the NOC has easy access to information on a given server that just alerted, and if you saved notes for that type of alert for that server, instructing Operators to collect certain information before they call you, that can save precious time and help you troubleshoot the issue. You might even let the NOC track incidents on a particular section of the wiki page for that server, so you can figure out if there is a pattern that you need to address.

YMMV on your organization’s needs – but this can be a very valuable tool to keep your business running efficiently. I have worked for teams that relied heavily on Wikis for documentation  - and saw first hand how this greatly improves access to information for our team. How you use it in your environment and how you adapt it to your requirements is entirely up to you – I only wanted to give you a few ideas on how you might put this to your advantage.

Which Wiki Engine Should I Use?

There are TONS of wiki engines out there: WikiPedia has a page (not frequently updated, alas) that lists some of them. Being the SQL Server guy that I am, I’m naturally inclined to use something powered by SQL in the backend. The most widely used wiki engine, developed by the Wikimedia Foundation, is a typical LAMP (Linux/Apache/MySQL/PHP) application. If you have experience with Apache, PHP and MySQL, you should use that. It’s updated very frequently, and it’s got a large community behind it.

Now, if you have SharePoint, you may or may not know that it comes with a wiki engine as well – and that’s detailed here.

Recently I used FlexWiki , which uses SQL Server as a backend. Being a SQL database on the backend, you back it up and keep it safe as the rest of your backups – just don’t use the same servers you’re trying to document to host the wiki engine’s database. If you lose that server, you lost access to the documentation about it.

Unfortunately, FlexWiki isn’t being actively maintained. But look at the list of Wiki software – you’re bound to find something that fits your needs.

No More Excuses – Get To It!

If you have no documentation/poor documentation. Go take care of it, now. Go create and/or reuse scripts that gather information about your environment. Be proactive about it – make sure those scripts get updated. Show your work to your peers and your boss. Feel confident that you have control of your environment, and rock on!

What Are Your Thoughts?

I’d love to hear what you’re doing in your environment as far as documentation goes. Will you leave some comments below? Also, if you feel like giving some feedback on the DBA Best Practices series, I’d appreciate it.

Until next time,

-Argenis

(I’m taking a break from my DBA Best Practices blog series to talk about this today – the series continues after the Holidays!)

I love to be proved wrong. Really, I do.

Recently we had a discussion in an distribution list where somebody asked whether a SQL Server 32-bit instance could address more than 4Gb of RAM when running on top of a 64-bit OS. One of the really smart guys in Microsoft SQL Server Support, Karthick P.K. [Blog|Blog|Twitter] replied “sure, just add more RAM and enable AWE – SQL will use that memory”. I was much convinced that this was incorrect, so I jumped in and said that AWE does nothing under those circumstances. After all, that is what I had read in the past in many different articles. I even had some SQL Server MVPs who have been around the block forever agreeing with me.

But the good news is that I was wrong.

AWE in fact does nothing on 64-bit instances of SQL Server. But on WOW64 - Windows on Windows 64-bit, the subsystem of Windows x64 that allows you to run 32-bit processes and instances of SQL Server, AWE does allow you to address memory above 4Gb. Let me show you.

Side note: AWE is now deprecated - removed in SQL Server 2012. The last version of SQL Server that supports AWE is SQL Server 2008 R2. Because of this, a SQL Server 2012 x86 instance won’t be able to enjoy over 4Gb of memory – even if running on WOW64.

I setup a vanilla VM in my lab with a 64-bit OS with 8Gb of memory. MSINFO32 looks like this on the VM:

I installed a 32-bit SQL Server 2008 R2 SP2 instance on the VM. @@VERSION looked like this:

I did not enable AWE. I did grant the “Perform Volume Maintenance Tasks” and “Lock Pages in Memory” privileges using secpol.msc (not pictured!). I wanted to see what memory consumption would look like without AWE enabled.

This screenshot shows the permissions granted to the SQL Server Service Account (NETWORK SERVICE, in this case) – used “whoami /priv”, a quick and dirty way to get the privileges of the service account using xp_cmdshell. Ugh, xp_cmdshell. It’s okay, it’s just a test.

I went ahead and created a test database with just one big table – big enough to use all the buffer pool memory on the instance when a size-of-data operation is executed on it. I did set Max Server Memory to 6Gb.

Here’s the CREATE table script:

CREATE TABLE [dbo].[testTable](
    [id] [int] IDENTITY(1,1) NOT NULL,
    [wideColumn] [nvarchar](4000) NULL,
        PRIMARY KEY CLUSTERED 
        (
            [id] ASC
        )WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]
) ON [PRIMARY]
GO
ALTER TABLE [dbo].[testTable] ADD DEFAULT (replicate(N'A',(4000))) FOR [wideColumn]
GO

Next, I went ahead and inserted 1M rows on the table:

INSERT INTO [dbo].[testTable] DEFAULT VALUES
GO 1000000

If you do the math very quickly, you will agree with me that this not so little table will be more than 6Gb in size. Let’s check anyway:

Cool! The table is large enough for the purpose of this test. But how does memory utilization look like after that 1M row insert?

That’s less than 4Gb. Next I turned on AWE, and restarted the instance. Then I queried the ERRORLOG to confirm that AWE was enabled:

Okay, since I had just restarted the instance my buffer pool was gone – needed to force a size of data operation on the large table. Rebuilding the clustered index did just fine (I know, that’s probably overkill). Then I checked sys.dm_os_process_memory again:

And there we go! A lot more than 4Gb being used now that AWE is in play. I had set Max Server Memory to 6Gb, as I detailed earlier.

It looks like there’s quite a lot of incorrect documentation regarding this, so I wanted to clear out any confusion – and learn something new along the way. Don’t you love SQL Server?

Happy Holidays and Happy New Year!

-Argenis

P.S.: If you are running IA64 (Itanium) - then this doesn't apply to you. AWE does not work at all on IA64.

This blog post is part of the DBA Best Practices series, on which various topics of concern for daily database operations are discussed. Your feedback and comments are very much welcome, so please drop by the comments section and be sure to leave your thoughts on the subject.

Morning Coffee

When I was a DBA, the first thing I did when I sat down at my desk at work was checking that all backups had completed successfully. It really was more of a ritual, since I had a dual system in place to check for backup completion: 1) the scheduled agent jobs to back up the databases were set to alert the NOC in failure, and 2) I had a script run from a central server every so often to check for any backup failures.

Why the redundancy, you might ask. Well, for one I was once bitten by the fact that database mail doesn't work 100% of the time. Potential causes for failure include issues on the SMTP box that relays your server email, firewall problems, DNS issues, etc. And so to be sure that my backups completed fine, I needed to rely on a mechanism other than having the servers do the taking - I needed to interrogate the servers and ask each one if an issue had occurred. This is why I had a script run every so often.

Some of you might have monitoring tools in place like Microsoft System Center Operations Manager (SCOM) or similar 3rd party products that would track all these things for you. But at that moment, we had no resort but to write our own Powershell scripts to do it.

Now it goes without saying that if you don't have backups in place, you might as well find another career. Your most sacred job as a DBA is to protect the data from a disaster, and only properly safeguarded backups can offer you peace of mind here.

"But, we have a cluster...we don't need backups"

Sadly I've heard this line more than I would have liked to. You need to understand that a cluster is comprised of shared storage, and that is precisely your single point of failure. A cluster will protect you from an issue at the Operating System level, and also under an outage of any SQL-related service or dependent devices. But it will most definitely NOT protect you against corruption, nor will it protect you against somebody deleting data from a table - accidentally or otherwise.

Backup, fine. How often do I take a backup?

The answer to this is something you will hear frequently when working with databases: it depends.

What does it depend on? For one, you need to understand how much data your business is willing to lose. This is what's called Recovery Point Objective, or RPO. If you don't know how much data your business is willing to lose, you need to have an honest and realistic conversation about data loss expectations with your customers, internal or external. From my experience, their first answer to the question "how much data loss can you withstand?" will be "zero". In that case, you will need to explain how zero data loss is very difficult and very costly to achieve, even in today's computing environments.

Do you want to go ahead and take full backups of all your databases every hour, or even every day? Probably not, because of the impact that taking a full backup can have on a system. That's what differential and transaction log backups are for.

Have I answered the question of how often to take a backup? No, and I did that on purpose. You need to think about how much time you have to recover from any event that requires you to restore your databases. This is what's called Recovery Time Objective. Again, if you go ask your customer how long of an outage they can withstand, at first you will get a completely unrealistic number - and that will be your starting point for discussing a solution that is cost effective.

The point that I'm trying to get across is that you need to have a plan. This plan needs to be practiced, and tested. Like a football playbook, you need to rehearse the moves you'll perform when the time comes. How often is up to you, and the objective is that you feel better about yourself and the steps you need to follow when emergency strikes.

A backup is nothing more than an untested restore

Backups are files. Files are prone to corruption. Put those two together and realize how you feel about those backups sitting on that network drive. When was the last time you restored any of those?

Restoring your backups on another box - that, by the way, doesn't have to match the specs of your production server - will give you two things: 1) peace of mind, because now you know that your backups are good and 2) a place to offload your consistency checks with DBCC CHECKDB or any of the other DBCC commands like CHECKTABLE or CHECKCATALOG. This is a great strategy for VLDBs that cannot withstand the additional load created by the consistency checks.

If you choose to offload your consistency checks to another server though, be sure to run DBCC CHECKDB WITH PHYSICALONLY on the production server, and if you're using SQL Server 2008 R2 SP1 CU4 and above, be sure to enable traceflags 2562 and/or 2549, which will speed up the PHYSICALONLY checks further - you can read more about this enhancement here.

Back to the "How Often" question for a second. If you have the disk, and the network latency, and the system resources to do so, why not backup the transaction log often? As in, every 5 minutes, or even less than that? There's not much downside to doing it, as you will have to clear the log with a backup sooner than later, lest you risk running out space on your tlog, or even your drive. The one drawback to this approach is that you will have more files to deal with at restore time, and processing each file will add a bit of extra time to the entire process. But it might be worth that time knowing that you minimized the amount of data lost. Again, test your plan to make sure that it matches your particular needs.

Where to back up to? Network share? Locally? SAN volume?

This is another topic where everybody has a favorite choice. So, I'll stick to mentioning what I like to do and what I consider to be the best practice in this regard. I like to backup to a SAN volume, i.e., a drive that actually lives in the SAN, and can be easily attached to another server in a pinch, saving you valuable time - you wouldn't need to restore files on the network (slow) or pull out drives out a dead server (been there, done that, it’s also slow!).

The key is to have a copy of those backup files made quickly, and, if at all possible, to a remote target on a different datacenter - or even the cloud. There are plenty of solutions out there that can help you put such a solution together. That right there is the first step towards a practical Disaster Recovery plan. But there's much more to DR, and that's material for a different blog post in this series.

Introduction

After the success of the “Demystifying DBA Best Practices” Pre-Conference that my good friend Robert Davis, a.k.a. SQLSoldier [Blog|Twitter] and I delivered at multiple events, including the PASS Summit 2012, I have decided to blog about some of the topics discussed at the Pre-Con. My thanks go to Robert for agreeing to share this content with the larger SQL Server community.

This will be a rather lengthy blog series - and as in the Pre-Con, I expect a lot of interaction and feedback. Make sure you throw in your two cents in the comments section of every blog post.

First topic that I’ll be discussing in this blog series: The thing of utmost importance for any Database Administrator: the data. Let’s discuss the importance of backups and a solid restore strategy.

Care to share your thoughts on this subject in the comments section below?

Today Jeffrey Langdon (@jlangdon) posed on #SQLHelp the following questions:

So I set to answer his question, and I said to myself: “Hey, I haven’t blogged in a while, how about I blog about this particular topic?”. Thus, this post was born.

(If you have never heard of Ghost Records and/or the Ghost Cleanup Task, go see this blog post by Paul Randal)

1) Do ghost records get copied over in a backup?

If you guessed yes, you guessed right. The backup process in SQL Server takes all data as it is on disk – it doesn’t crack the pages open to selectively pick which slots have actual data and which ones do not. The whole page is backed up, regardless of its contents.

Even if ghost cleanup has run and processed the ghost records, the slots are not overwritten immediately, but rather until another DML operation comes along and uses them.

As a matter of fact, all of the allocated space for a database will be included in a full backup.

So, this poses a bit of a security/compliance problem for some of you DBA folk: if you want to take a full backup of a database after you’ve purged sensitive data, you should rebuild all of your indexes (with FILLFACTOR set to 100%). But the empty space on your data file(s) might still contain sensitive data! A SHRINKFILE might help get rid of that (not so) empty space, but that might not be the end of your troubles. You might _STILL_ have (not so) empty space on your files!

One approach that you can follow is to export all of the data on your database to another SQL Server instance that does NOT have Instant File Initialization enabled. This can be a tedious and time-consuming process, though. So you have to weigh in your options and see what makes sense for you. Snapshot Replication is another idea that comes to mind.

2) Does Compression get rid of ghost records (2008)?

The answer to this is no. The Ghost Records/Ghost Cleanup Task mechanism is alive and well on compressed tables and indexes. You can prove this running a simple script:

CREATE DATABASE GhostRecordsTest
GO

USE GhostRecordsTest
GO

CREATE TABLE myTable (myPrimaryKey int IDENTITY(1,1) PRIMARY KEY CLUSTERED,
                      myWideColumn varchar(1000) NOT NULL DEFAULT 'Default string value')
                       
ALTER TABLE myTable REBUILD PARTITION = ALL WITH (DATA_COMPRESSION = PAGE)
GO

INSERT INTO myTable DEFAULT VALUES
GO 10

DELETE myTable WHERE myPrimaryKey % 2 = 0

DBCC TRACEON(2514)

DBCC CHECKTABLE(myTable)

TraceFlag 2514 will make DBCC CHECKTABLE give you an extra tidbit of information on its output. For the above script: “Ghost Record count = 5”

Until next time,

-Argenis

I took the MCM lab a couple of weeks ago and passed. I still don’t believe it - it’s been quite the journey.

There have been many decisive moments in my career; one of those was back in 2010 when I decided I would work on obtaining the certification after Brent Ozar blogged rather extensively about his experience with the program. I progressively worked on aspects of SQL Server that were out of my comfort zone – and that paid off. But simply put, I have the cert - but still don’t consider myself a true Master. And I probably never will.

I might be okay in certain areas of SQL, maybe even good in others - but I don’t master them all. I intend on getting better on those areas, but that won’t happen overnight.

SQL Server is a mammoth of a product. It’s actually a suite of products – and the MCM certification focuses on only one aspect of it: the Database Engine. No SSAS, no SSIS, no SSRS.

So you might be wondering: Is it worth it then? Why work so hard on obtaining a certification that calls you “Master” when it doesn’t test your skills on every single feature of SQL Server?

To me, it was. And then some.

The fact that I value most after going through the program? Knowing exactly what I don’t know. In other words, looking at a problem and knowing that I won’t be able to solve it right away, yet still I will have an idea of where to begin digging for information to conquer it.

You see, when I decided to go for the MCM I would have probably ranked myself a 6/10. So I went and trained myself on those areas where I felt I wasn’t strong enough. I read blogs, I watched videos, I attended SQLSkills immersion events (lucky me!), I went to SQLSaturdays, and even a couple of PASS Summits – but all of that was theory, not a whole lot of practice. And when you get to the MCM Lab, you better had practiced. You better had familiarized with that particular feature you’re being challenged with – at least a tiny bit. Learning from BOL on the fly during the MCM Lab is probably just going to be a waste of time – trust me on that one.

The other thing I learned along the way? The power of networking. If you are a SQL Server professional, in any function – DBA, Dev or what have you – allow yourself a couple of minutes to brag about what you do: our community is awesome. Better than anything I’ve seen out there. I haven’t met a single person in the SQL community that’s not willing to help a colleague.

And that’s precisely what helped me prepare the most for the MCM: helping others. Perhaps that will help you get there as well?

Cheers,

-Argenis

Edit: I have confirmed that this is still valid for the RTM version of SQL Server 2012.

If you recall one of my previous blog posts, titled Think Your Windows Administrators Don’t Have Access to SQL Server 2008 by Default? Think Again I exploited the fact that NT AUTHORITY\SYSTEM was granted membership to the sysadmin server role by setup in SQL Server 2008 R2 and below to gain access to a SQL instance to which I had no access, since as Administrator on the box I could launch a cmd session as NT AUTHORITY\SYSTEM with Sysinternals’ psexec utility.

My friend and SQL Server MVP Jorge Segarra [blog|twitter] correctly pointed out shortly after that NT AUTHORITY\SYSTEM is not a member of the sysadmin server role in SQL Server 2012, codename Denali. And as of Release Candidate 0 for this version, this holds true.

What also holds true as of RC0 is that the Service SID for a number of services (at least three, the SQL Engine itself, SQL VSS Writer and Winmgmt) are members of the sysadmin role. And so in this post I’d like to demonstrate that it is possible to exploit one of these services' level of access to hop onto a 2012 instance as sysadmin.

The target: a named SQL instance called “DENALI_RC0” on one of my desktop PCs. Having dropped my login on SQL, when I try to logon to the instance I get the usual message:

I picked a service to become “the victim”. The SQL VSS Writer service seemed to be a good candidate: innocuous enough. If I stop it and restart it, no big deal.

I launched regedit and browsed to HKLM\SYSTEM\CurrentControlSet\services\SQLWriter - this is what I saw:

Now being an Administrator of this PC as I am, I went ahead and renamed sqlwriter.exe to sqlwriter.exe.orig, and put a copy of SQLCMD.EXE on C:\Program Files\Microsoft SQL Server\90\Shared.

Then I renamed SQLCMD.EXE to sqlwriter.exe.

Obviously kicking off the SQL VSS Writer service was not going to do anything – just error out:

So I replaced the ImagePath for sqlwriter on the registry with this:

"C:\Program Files\Microsoft SQL Server\90\Shared\sqlwriter.exe" -S CSHQAFERNANDEZD\DENALI_RC0 -E -Q "CREATE LOGIN [CORP\Argenis.Fernandez] FROM WINDOWS; EXECUTE sp_addsrvrolemember @loginame = 'CORP\Argenis.Fernandez', @rolename = 'sysadmin'"

And now I kick off the sqlwriter service again, expecting it to error out…but with a nice side effect.

Sure enough, launched SSMS 2012 and was able to login. And guess what, my login has sysadmin privileges.

And so I’m sure some of you have already yelled “SECURITY HOLE!!!!” by now – yeah, to a degree…but remember kids, if you’re a local Administrator on the box, you already own the box. Very little applications like SQL Server can do to protect themselves from a “rogue” Admin. Maybe a few adjustments to the security model for Windows’ SCM (Service Control Manager) are needed here, but I’ll let you decide on that.

Cheers,

-Argenis